Implications of short-term Cargo Collapses on European

Aaron B. Scholz°, Benedikt Mandel*, Axel Schaff er+

° Institut für Wirtschaft spolitik und Wirtschaft sforschung (IWW), Karlsruher Institut für Technologie (KIT), Germany * MKmetric, Gesellschaft für Systemplanung mbH, Karlsruhe, Germany + Institut für die Entwicklung zukunft sfähiger Organisationen, Universität der Bundeswehr München, Germany

Keyword: air freight, cargo transport, network analysis, input-output analysis, risk analysis.

1. Motivation The structure of the paper is as follows. Chapter 2 defi nes, in a fi rst step, the air cargo network as a matrix that accounts In the aftermath of Iceland’s Eyjafjalljökull volcano’s for freight volumes shipped from i to any other eruption the Airspace in Germany and much of Europe was airport j. The matrix comprises 291 European airports, fi ve closed for days in mid April 2010. The major disruptions of artifi cial airports, representing the rest of the world, and each the fl ight schedule, caused by the giant ash cloud emitted by airport’s hinterland. Flows assigned to the functional airports the volcano, left thousands of passengers stranded at airports derive from the aggregated fl ows for the major airports in all over the world and turned out to be particularly damaging the regions of North America, South America, Africa, Asia- for European airlines. According to estimations of the Pacifi c and Rest of Europe. International Air Travel Association (IATA) the widespread The derived matrix forms the starting line for the measurement closure of European airspace in April came along with a loss of each airport’s importance performed in chapter 3. For this in revenues of around €1.26 billion. purpose the air cargo network is interpreted as an input- Clearly, economic damages could easily exceed airlines’ output system with the 296 x 296 matrix as the intermediate losses. Since Europe’s fragile economies are still recovering part and the in- and outfl ows from and to the corresponding from 2008’s and 2009’s fi nancial crises, any major disturbance hinterlands as primary inputs and fi nal demands respectively. of the production process might hamper the economic Following the principles of input-output analysis, the next turnaround. This is particularly true for industries that rely step foresees to calculate different linkage indices in order to heavily on air cargo such as Europe’s car manufacturers. reveal each airport’s importance in the network. BMW, for example, was forced to halt production due to Chapter 4 quantifi es the impacts of single (and multiple) missing electronic components. The example highlights airport closures on the entire European air cargo market. the dependency of manufacturer on a smooth running air Such closures might accrue from natural disasters, terrorist transport system. A system that is highly dependent on the attacks or technical breakdowns and might have far-reaching whole network of European airports and therefore very implications also for not directly affected airports. Therefore, sensitive for any kind of disturbances.1 the input-output philosophy is applied again. With regard to the frequent occurrence of temporary Finally, chapter 5 presents the main conclusions drawn from disturbances, such as severe natural events or potential this study. terrorism, and having in mind the strong interdependence of economic production, trade and (air) freight transport, the main objective of the paper is to identify the importance of 2. Air cargo matrix each airport for the entire (European) air cargo network and to model the impacts of disasters on Europe’s airports. The set up of the European air cargo matrix requires 1 Despite signifi cant impacts of these temporary disturbances, there is comprehensive, homogenous and reliable data. These hardly any evidence that they shall affect the major trend in air car- requirements are fulfi lled by data that derive from the go development. In the past decades, this trend was characterized by research project WORLDNET (2009a, 2009b).2 Within the growth rates well above worldwide GDP growth. In fact, freight tonnes scope of this project air cargo is incorporated into a European transported by air rose by more than 5% per year between 1995 and 2007. Thus, air cargo volumes outpaced the growth of global GDP be- transport network model. This, in turn, allows for setting up tween 1.5 and 2 times. Things changed dramatically in the second half a fl ow matrix on the given regional for the year 2005 of 2008. The worldwide decline of industrial production and a strong covering air cargo within Europe and between Europe and reduction of international trade volumes hit the logistics business in general and the inter-continental/long-distance air cargo business in 2 WORLDNET is a Framework 6 research project under the Scientifi c particular. The slump has affected all European airports but their mag- Support to Policies (SSP) initiative of the European Commission, Di- nitude differs signifi cantly. rectorate General Energy and Transport.

Vol. 37 (3-4) 2011 Der öff entliche Sektor - Th e Public Sector 39 A.B. Scholz, B. Mandel, A. Schaff er

Source: WORLDNET, 2009a Fig. 1. Scheme of the developed approach for constructing the air cargo matrix

the world. Thus, the model copes with the dominance of The iterative calibration process (assignment) grounds on a intercontinental fl ow patterns - the underlying regionalization route choice model. In order to minimize the residuum of a of WORLDNET, with about 500 traffi c zones representing linear system of equations, the model assigns fl ows of the the world outside Europe and more than 1500 for Europe start-up matrix to concrete routes (Furness algorithm). The (NUTS3-level) form the base for an adequate detail. Figure linear-system consists of one line per air link and one column 1 illustrates the developed scheme of the WORLDNET per origin-destination pair. The iterative process terminates approach for air cargo fl ows. when the deviation between statistics and link loads as well as between start-up matrix and adapted matrix have come to The main approach consists of four major sequences a minimum (WORLDNET, 2009b). (WORLDNET, 2009a, 2009b). The developed air cargo matrix shows the cargo fl ows in 1. Build up an air cargo network, combining air links tonnes for each pair of regions covered within WORLDNET. between airports and feeder links between regions and At least one of these regions has to be situated in Europe airports (nodes and links), (the matrix covers all cargo fl ows to, from or within Europe). 2. Enrich the build-up air cargo network with actually Cargo fl ows between two regions outside Europe are not measured cargo volumes from statistics (airport and fl ow covered, irrespective the routing touches Europe or not. statistics), Figure 2 presents the cargo volumes in tonnes for the year 3. Create from the available statistics a start-up freight 2005. The total air cargo fl ows to/from and within Europe matrix for air cargo, including also domestic fl ows sum up to 11.25 million 4. Run an iterative calibration process to minimize the total tonnes for the year 2005. Three fourth of this volume has sum of deviations between assigned (model results) and been carried on intercontinental relations, while 20% were measured (observed) link loads as well as between start- international, intra-European transport and less than 5% up cargo fl ows adjusted according to the link statistics on were of domestic nature. country-level. Air cargo to or from Asia forms about 40% of all freight fl ows of this mode, while cargo to or from America makes Data sources which have been used for the air cargo matrix about 25% of the total amount. Cargo to from Africa brings include trade statistics (EU Trade Transport Data) to identify another 7% of the cargo volume, while the fl ows concerning the air mode for extra-EU trade fl ows, transport statistics Australia/ Oceania are minor building a share of just 3% (e.g. Eurostat, national statistics) to gain information on the of all air cargo fl ows. The remaining quarter is the intra- intra-European air cargo fl ows and socio-economic data (e.g. European air cargo demand. Eurostat, CIA World Factbook) for the purpose of matrix calibration. Considering the country specifi c volumes, the top fi ve

40 Der öff entliche Sektor - Th e Public Sector Vol. 37 (3-4) 2011 Implications of short-term Cargo Collapses on European Airports

Source: WORLDNET, 2009a Fig. 2. Cargo Volumes by European country

countries in Europe are Germany (3.2 mill. tonnes), United to calculate the airports’ degree of interdependency and Kingdom (2.1 mill. tonnes), France (1.7 mill. tonnes), importance. The Netherlands (1.1 mill. tonnes) and Belgium (1.0 mill. tonnes), which makes about 80% of all air cargo carried to, from or within Europe. 3. Importance and interdependency of Euro- The WORLDNET matrix, which covers all freight fl ows to, pean airports from or within Europe, can be considered the core of the matrix applied for this study. However, to set up a complete air cargo matrix, the fl ows among the functional airports and 3.1. Air cargo input-output table the fl ows between the functional airports and their hinterland In order to identify the importance and interdependency of have to be estimated in a last step. European airports, the study at hand follows the input-output With regard to the missing fl ows among the artifi cial airports technique. Traditional input-output tables provide detailed the “World Intercontinental Air Trade Forecast” which information of industries inputs and outputs in matrix form. grounds its forecasts for the period from 2006 to 2010 on 2005 The core of the tables is the intermediate quadrant that gives data gives suffi cient information (Merge Global, 2006). For an insight into the industries’ linkages. Under ceteris paribus the estimation of the fl ows from and to the hinterland average restrictions, the input-output representation thus allows European-Airport-Hinterland-Coeffi cients are calculated for analyzing the dependency of each industry on all other and then applied to the aggregated artifi cial airports (see industries. Since the tables also account for the industries annex for a detailed description of the procedure).3 production of fi nal goods and their absorption of primary inputs, they present a rather complete picture of an economy. The outcome of the described steps fi nally allows setting up This also implies that total inputs – generally measured in the complete input-output matrix for air cargo. This complete monetary units equal total outputs for each industry. If the matrix structure is needed for accomplishing the analyses tables are used for analytical purposes, the technology is assumed to be constant. While this is reasonable for short, 3 Clearly, this rather rough method hardly yields exact results. But since and in some cases medium term analysis, the tool is (in its hinterland fl ows for the rest of the world airports hardly affect our analysis, which focuses on the identifi cation of European airports’ rela- static version) less appropriate for the simulation of long tive importance, we prefer to go on with an educated guess rather than term effects. empty cells.

Vol. 37 (3-4) 2011 Der öff entliche Sektor - Th e Public Sector 41 A.B. Scholz, B. Mandel, A. Schaff er

Table 1. General Structure of the input-output table for air cargo fl ows

Source: Vosen, 2011

For the study at hand, the input-output philosophy is airports (including 5 rest of the world airports) to airport j. transferred to the air cargo market. For this purpose the n airports replace the industries as consumers of inputs and Each row sum (  xij ) displays the outfl ows of airport i to producers of outputs. Both inputs and outputs (henceforth j1 infl ows and outfl ows) are measured in tonnes. The point of all airports (including 5 rest of the world airports). departure for the analysis is a set of 291 European and 5 artifi cial airports (representing the rest of the world) where 3.2. Measure of airport interdependency and importance air cargo fl ows occurs in a certain period of time (base year A smooth running air cargo network requires that each airport 2005). Table 1 displays the general structure of the input- receives the infl ows from other airports and its hinterland output matrix for air cargo fl ows. according to the planned scenario. In case this requirement Cargo fl ows between 291 European airports and between cannot be met, as it was the case in the aftermath of the recent these airports and 5 world regions4 (blue shaped) are eruption of Iceland’s Eyjafjalljökull volcano, the production incorporated into the matrix. Furthermore, feeder services of airports and clients in the respective hinterland might be to (green shaped) and from (orange shaped) the European severely limited. The production stop of BMW in Munich airports are also considered (“Hinterland”). due to missing parts shipped by air is only partly due to the Following the input-output notation, the blue-shaped part closure of . Equally important was the closure can be considered the intermediate matrix. A typical element of main European hub airports, which receive freight from all over the world and further distribute the goods to their xij denotes the total freight tonnes shipped from airport i to airport j. The rows display the outfl ows of airport i whereas the fi nal destination. Furthermore Munich manufacturers receive columns illustrate the airports’ infl ows. The row “Hinterland” their air cargo not only from Munich airport but also from displays the air cargo outfl ows from airport i to its hinterland other nearby airports. Figure 3 depicts the combined cargo (by any other mode). In input-output notation this vector catchment of some selected airports addressing the market defi nes airport i’s fi nal demand. The column “Hinterland” share they can attract from the regions. In fact the spatial (the primary input vector in the traditional tables) displays coverage shows how strong industries are affected in case air cargo infl ows to airport j from its hinterland (by any other of an airport closure. So the industry around Munich would mode). have been affected as well to a level of 10% despite the airport n MUC would have been operating but the depicted ones Each column sum (  xij ) displays the infl ows from all would have been closed. In consequence risk assessment of i1 production lines should consider this dependency. 4 The fi ve aggregated World Regions are: Africa (AF), Asia Pacifi c (AP) Clearly the hinterland increases with airport size, which (Asia and Oceania), North America (NA), South America (SA), Rest of Europe (not considered in the detailed analysis of the WORLDNET can certainly be considered a fi rst indicator of an airport’s project). importance. Figure 4 shows the 20 biggest European airports

42 Der öff entliche Sektor - Th e Public Sector Vol. 37 (3-4) 2011 Implications of short-term Cargo Collapses on European Airports

Source: authors’ own representation Fig. 3. Hinterland of the airports Amsterdam, Charles de Gaulle, Luxembourg, Vienna and Zurich

Frankfurt 1116 Amsterdam 1016 Charles de Gaulle 857 Heathrow 763 422 Cologne / Bonn 355 Luxembourg 335 Milan - Malpensa 265 Munich 205 204 Zurich 200 Liege 177 London Stansted 172 Copenhagen 155 East Midlands 140 London Gatwick 139 Manchester 139 Fiumicino 125 Prague 118 Baku 113

0 200 400 600 800 1000 1200 1000 t Source: authors’ own representation Fig. 4. Cargo volumes shipped to other airports

Vol. 37 (3-4) 2011 Der öff entliche Sektor - Th e Public Sector 43 A.B. Scholz, B. Mandel, A. Schaff er in terms of freight volumes shipped to other airports. The Taking into account the Ghosh inverse matrix, equation (4) volumes account for fl ows to the rest of the world airports describes the complete airport cargo network: but do not include fl ows to the respective hinterland. (4) x = (I – A)-1 f In terms of absolute fl ows, the closure (or limited operation) of Amsterdam airport has more severe impacts for the network x: n-element vector of airports’ total infl ows operability compared to a closure (or limited operation) of (= airports’ total outfl ows) Paris Charles de Gaulle. However, the degree of a network’s inoperability is further determined by the (supply driven) f: n-element vector of airports’ infl ows from forward linkages of the affected airport(s). the corresponding hinterland (other infl ows)

The fi rst step to identify the airports’ forward linkage is the Furthermore, the Ghosh inverse matrix enables the calculation calculation of the output-coeffi cients a : ij of input multipliers in line with equation (5) (West, 1988):

xij n (1) aij  (5) x Li  bij i j1 The coeffi cient measures the cargo volumes that are further In the context of the air cargo input-output system, the shipped from airport i to airport j (outfl ows) in relative multiplier measures the cumulated effect on total outfl ows 5 terms to the total infl ows received by the airport. Following of all airports that come along with a unit change in the one basic assumption of the input-output analysis, the aij infl ows of airport i. Thus, Li measures how much more (or are considered constant in the short term which is not an less) cargo is further shipped to any other airport when an unrealistic scenario (e.g. see the Eyjafjalljökull volcano’s additional unit (or a unit less) of cargo arrives at airport i. eruption). A coeffi cient of 0.4 implies that for each 100 Cargo can arrive either from airport to airport relations or tonnes of cargo arriving at airport i (from other airports or the from the airport’s hinterland. A multiplier for airport i of airport’s hinterland), 40 tonnes are further shipped to airport 2.50 means that in case of a sudden drop of infl ows by 1,000 j. If, for some external reasons, infl ows received by airport i tonnes to this airport, the transported tonnes in the whole are reduced by 50%, or if the airport receives all infl ows but network is reduced by 2,500 units where 1,500 units occur at can, due to internal defects, process only 50% of the volume, other airports (transfer freight). airport j just receives 20 tonnes. A high multiplier value identifi es a strong forward linkage In a second round, the reduction of 20 tonnes in the infl ows and therefore points to a relatively high importance of the for airport j causes further disruption of the system. Assume airport as a supplier of intermediate goods. The limited the as planned scenario of airport j foresees to ship a total operability of such an airport, or in case of an extreme event of 100 tonnes to other airports, production is now limited to the closure, has severe consequences for the whole network. 80 tonnes. This effect continues, to an ever minor extent, in Clearly, major hub airports can be assumed to come up with all subsequent rounds. Thus, the estimation of the cumulated high multipliers. In contrast, we expect small multiplier for effect follows an iterative process according equation (2): airports at the end of the logistic chain. (2) B = I + A + A2 + A3 + A4 + ... Table 2 shows the 20 airports with the highest multiplier but also gives some examples for well-known airports with I: Unity matrix medium and small multipliers.7 The results confi rm the expectations in most instances. Major a11 ... a1n    cargo hubs, such as Frankfurt (FRA), Amsterdam (AMS), A:   Paris (CDG) and London Heathrow (LHR) are among the top   20. Milan Malpensa International (MXP) and Luxembourg an1 ... ann   (LUX) are just behind. In addition Madrid, Brussels and (The A-matrix is often referred to as Ghosh matrix.6) Cologne / Bonn and Rome show over-average multipliers, and lead the group of airports with medium multipliers. The result of the iterative process given by equation (2) can Since the multipliers are independent from the absolute fl ows alternatively be calculated according equation (3): shipped by the airport, results are not driven by the airports’ size but their (forward) interconnectedness. This explains (3) B = (I-A)-1 why smaller hubs of larger airlines (e.g. SAS/Copenhagen, Swiss/Zurich, /Munich, Cargolux/Baku) come up b11 ... b1n  with similar or sometimes even higher multipliers compared   to the leading European cargo hubs. Finally, some airports, B:     such as Kristiansand airport Kjevik or Pisa emerge in the top b ... b  20 list by surprise. However, these airports have rather strong  n1 nn (The B-matrix is often referred to as Ghosh inverse matrix.) 7 The calculation of multipliers is based on the full cargo input-output table. However, since the focus of the presented study is on air cargo transport in Europe, airports are only considered for further discussion, 5 Note that total infl ows equal total outfl ows for any airport. if a minimum freight volume of at least 1000 t per year is processed at 6 A detailed review and interpretation of the Gosh-matrix can be found at the respective airport. This restriction limits the analysis of airports’ Dietzenbacher (1997). importance to a total of about 180 airports.

44 Der öff entliche Sektor - Th e Public Sector Vol. 37 (3-4) 2011 Implications of short-term Cargo Collapses on European Airports

Table 2. Forward linkages of European airports Airport Multiplier Airport Multiplier High multipliers Medium multipliers Rovaniemi 2.81 Madrid 2.41 Baku 2.75 Brussels 2.34 London Gatwick 2.73 Cologne / Bonn 2.33 Coventry 2.65 Rome Fiumicino 2.33 Frankfurt 2.64 Innsbruck 2.21 Paris Charles de Gaulle 2.62 Berlin 2.18 Pisa 2.62 Dublin 2.16 Amsterdam 2.61 Split 2.16 Belgrade 2.61 Athens 2.16 Riga 2.60 Warsaw 2.15 Copenhagen 2.60 Vienna 2.07 Bucharest 2.58 Turin 2.06 London Heathrow 2.57 Low multipliers Kjevik 2.57 Friedrichshafen 1.73 Paris Orly 2.56 Klagenfurt 1.73 Munich 2.55 1.71 Strasbourg 2.53 Jersey 1.61 Basel Mulhouse 2.52 Heraklion 1.59 Stockholm 2.52 Cagliari 1.53 Zurich 2.51 1.44 Source: authors’ own representation

linkages with major hubs (Copenhagen and Amsterdam in n case of Kjevik, Rome Fiumicino and Malpensa International bij in case of Pisa) and benefi t indirectly from the strong  j1 (6) L   interconnectedness of those airports. i i n n bij We further expected small multipliers for airports at the end i11j of the logistic chain. Again, the results generally strengthen this hypothesis, since peripheral airports in touristic regions : weight (without major manufacturing industries) largely form the i group of airports with smallest forward multipliers. Clearly the linkage is rather sensitive to the chosen weight. Although the calculation of multiplier is technically The presented study two different weights are proposed. independent of the airport size, a relation can still exist. An First, the attached weight just equals the absolute cargo airport’s strong interconnectedness could for example add volumes (in tonnes) shipped from airport i to other airport. to its attractivity as a cargo hub and therefore enhance the Since the multipliers are rather similar for the biggest processed freight volumes over time. Figure 5 partly supports airports, the linkage relies heavily on the airports’ size and this reasoning, as airports’ total outfl ows (without fl ows to hardly any change compared to the ranking based on tonnes the corresponding hinterland) increase with the multipliers. only can be expected. The second weight is also based on At the same time the trend is not very strong. Indeed low and the processed cargo volumes, but instead of the absolute high multipliers occur for any airport size. fl ows their logarithmic value is taken into account. Thus, the The fi nal step foresees the identifi cation of airports’ attached weights still increase with the absolute fl ows but importance by taking into account airport size and forward at a much lower rate. Table 3 shows the airports with the linkage. For this purpose a weighted (and this time highest forward linkages for both alternatives. normalized) linkage index is defi ned in the following way: In case the weights are defi ned by the absolute fl ows, the rankings hardly change compared to the ranking based on

Vol. 37 (3-4) 2011 Der öff entliche Sektor - Th e Public Sector 45 A.B. Scholz, B. Mandel, A. Schaff er

Source: authors’ own representation Fig. 5. Airports’ total outfl ows (in 1000t logarithmic scale) and forward multipliers

Table 3. Weighted forward linkages with diff erent weights Airport Weighted delta Airport Weighted delta linkage rank linkage rank Frankfurt 22.51 0 Frankfurt 1.86 0 Amsterdam 20.25 0 Amsterdam 1.82 0 Paris Charles de Gaulle 17.14 0 Paris Charles de Gaulle 1.81 0 London Heathrow 14.99 0 London Heathrow 1.76 0 Brussels 7.51 0 London Gatwick 1.64 +11 Luxembourg 6.33 +1 Baku 1.61 +14 Cologne / Bonn 6.32 -1 Luxembourg 1.59 0 Milan - Malpensa 5.00 0 Munich 1.58 +1 Munich 4.00 0 Copenhagen 1.57 +5 Zurich 3.82 +1 Milan Malpensa 1.56 -2 Madrid 3.75 -1 Zurich 1.55 0 London Stansted 3.22 +1 Brussels 1.53 -7 Liege 3.11 -1 Cologne / Bonn 1.51 -7 Copenhagen 3.08 0 London Stansted 1.49 -1 London Gatwick 2.91 +1 Madrid 1.49 -5 Baku 2.36 +4 Prague 1.47 +3 East Midlands 2.30 -2 Stockholm 1.46 +5 Prague 2.25 +1 Paris Orly 1.44 +10 Manchester 2.23 -2 Liege 1.40 -7 Rome Fiumicino 2.21 -2 Milan Orio al Serio 1.40 +4 * delta rank: changes of ranking compared to the ranking based on cargo volume only (fi gure 4)

Source: authors’ own representation

46 Der öff entliche Sektor - Th e Public Sector Vol. 37 (3-4) 2011 Implications of short-term Cargo Collapses on European Airports cargo volume only (compare fi gure 4). Some airports switch The approach grounds on the relationship of the input-output position, but with the notable exception of Baku airport no philosophy where input and outputs are linked together in the signifi cant changes occur. In fact all of the biggest 20 airports following way (introduced in chapter 3): in terms of cargo volume reappear in the top 20 list of airports with highest weighted forward linkages. x = (I – A)-1 f In contrast, the ranking changes signifi cantly, if the weights are based on the logarithmic values. The airports of Baku, x: n-element vector of airports’ total infl ows London Gatwick, Paris Orly, Copenhagen and Stockholm (= airports’ total outfl ows) clearly rise in importance. At the same time, the airports of f: n-element vector of airports’ infl ows from East Midlands, Brussels, Cologne / Bonn, Liege and Madrid the corresponding hinterland (other infl ows) lose 5 or more positions due to their comparatively smaller forward connectedness. The airports of East Midlands, The implemented approach and its three operation steps are Manchester and Rome Fiumicino even drop out of the top introduced in the following: 20 list. They are replaced by the airports of Stockholm, Paris Orly and Milan Orio al Serio. Step 1 (recalculation of the intermediate matrix’s elements Due to their much higher volumes compared to all other x ): airports, the biggest cargo hubs (Frankfurt, Amsterdam, ij Paris CDG and London Heathrow) remain at the top of the  The disruption share (ρi) of all engaged airports of the ranking, no matter which weight is attached. European air freight system is to be determined fi rst (disaster scenario defi nition). A disruption share of i.e. 0.4 implies that after the disaster 40% of freight tonnes 4. Implications of airport closures on the entire can still be handled at airport i. The remaining 60% air cargo network cannot be operated from/to airport i which might be due to closure due to damage, ice, etc., increased safety distances between aircrafts, de-icing of aircrafts, 4.1. Model defi nition and development etc. The introduced forward linkages give a fi rst impression  The intermediate output-coeffi cient matrix A needs to be on Europe’s central and most important airports. In a next adapted based on the developed disaster. For each engaged step the impacts of network disruptions, such as airport airport i its row and column values xij are multiplied by or air space closures are quantifi ed by applying an input- its disruption share ρi. The adapted values x’ij indicate the output model. The idea of the model is leant on the work total freight tonnes which can be transported from the of Jiang and Haimes (2004) who developed a framework disrupted airport i to airport j caused by the disaster. based on Leontief’s notation of input-output and applied it for an economic risk assessment. Jiang and Haimes (2004) Step 2 (calculation of the “disaster matrix” A*): use a steady-state approach which describes primarily the long-term, equilibrium relationship for the participating  The adaptations of the matrix’s elements (x’ij) also subsystems. change the structure of the original input-output table, The present subsystem is the European air freight market and the A-matrix. Therefore, a new “disaster matrix” A* is the model’s main features are as follows: required which results from these adaptations and which is determined as introduced in chapter 3 for the original  Airport (air space) closures (or disruptions) result in A-Matrix. changes in the freight fl ows between the affected airport(s)  Furthermore, a new Ghosh-Matrix as well as a new and their connected airports (hinterland transports keep Ghosh-Inverse is determined which are based on A* unchanged) (“disaster matrix”). The new Ghosh-Matrices describe  Freight fl ow changes lead to an adaptation of the the “new” interdependencies and interconnectedness intermediate matrix, the A matrix where a typical element between the system’s airports which are limited by the

xij denotes the total freight tonnes shipped from airport i present disaster. to airport j  Cascading effects caused by the interdependencies and Step 3 (recalculation of the steady-state) interconnectedness of the European air freight market intensify the impacts of the initial airport closures and  The impact of the disaster for each airport as well as for will change the overall air freight network size the entire airport system is now recalculated by applying the new Ghosh-Matrices to the original infl ows of the Disruption at the airport level is defi ned as the remaining system, the Hinterland transports. capacity (percentage of the daily average capacity, disruption share) which still can be operated at the airport(s) after x* = (I – A*)-1 f the disaster. Disasters might accrue from natural disaster, terrorist attacks or technological problems which impact the Finally, the results (x*) can then be compared with the maximum capacity at the airport(s). original, non disaster affected, results (x) of every airport.

Vol. 37 (3-4) 2011 Der öff entliche Sektor - Th e Public Sector 47 A.B. Scholz, B. Mandel, A. Schaff er

Deviations indicate the disruption level at airport i which is freight tonnes of approx. 18% even though around 80% of due to the interdependency and interconnectedness of the air airports are not directly affected by the winter storm. freight system where also not directly affected airports do The most impacted airports are the airports which are directly perceive the disaster signifi cantly. affected by the winter storm because capacity reductions at these airports are incorporated into the model for the air side 4.2. Case-study application exogenously9 (intermediary matrix). The top 10 affected airports which are not located in the winter storm regions are The developed model can be applied to airport closures presented in Table 5 for both scenarios. It becomes obvious and capacity reductions at airports as well as to closures that especially the interconnectedness with the major air and capacity reductions of entire air spaces in Europe. A freight gateways impact the disruption levels of not directly recurring challenge to Europe’s air freight system is capacity affected airports (and regions). Therefore, Polish, Italian, constraints due to winter storms (including heavy and abrupt and Greece airports are the top affected airports of scenario snowfall). Therefore, the vulnerability of the entire air freight 1 because international gateways to the major air freight system as well as the direct and indirect impacts of winter market, such as Asia and North America, are either not storms on airports will be analyzed in detail. Airports which located in these countries or freight is to a signifi cantly share are not directly but indirectly affected by the winter storm transported by feeder services to the major cargo airports will be elaborated in detail (cascading effect). in central Europe (e.g. AMS, CDG or FRA). Nevertheless, The applied scenario is leant on the situation in December the top 10 airports which are primarily affected by scenario 2010 where a heavy snow storm affected airports in Germany, 1 are all of secondary importance for the entire European the Benelux, Great Britain and in North France. Besides a air freight market that only the fi rst order impacts do count large number of smaller airports also the European air freight for this scenario (all Austrian, Czech and German airports gateways have been affected by the storm signifi cantly, such are constraint by the winter storm). A different picture can as Amsterdam Airport Schiphol (AMS), Paris Charles-de- be observed for scenario 2. The overall daily freight tonnes Gaulle (CDG), Frankfurt International (FRA) and London are reduced by approx. 18% which is already a signifi cant Heathrow (LHR). The very high importance of the gateways reduction for the air freight system. Furthermore, approx. for the European air freight system as already introduced 20% of European airports are directly constraint by the by the forward linkages and therefore the high dependency winter storm including the air freight gateways and fi nally, of secondary airports on their operability underline the more airports are signifi cantly affected by the storm caused importance of such a scenario for the gateway airports but by secondary and tertiary effects. Especially, British and particularly for the dependent secondary airports. French airports are now present in the top 10 which can The scenario “winter storm in Europe” is divided into two be explained by the fact that only some British and French sub-scenarios which represent the spread of the winter storm airports are defi ned as directly affected by the winter storm. over time8. Figure 6 illustrates the situation and colours Due to the strong interdependencies and interconnectedness the affected regions of the sub-scenarios. Scenario 1 (light between airports of the same country and especially within shade) analyzes the beginning of the winter storm: First Great Britain caused by its island position, also other airports disturbances of the airlines’ fl ight schedules are observed are indirectly affected by the winter storm. As for scenario and one of the major air freight gateways, namely FRA, is 1 it should be noted that the top 10 airports (by disruption already affected by the winter storm. Scenario 2 (dark shade) level) are only secondary airports but the entire system tightens the situation and the storm is spread over central disruption and especially the (partly) inoperability of the air Europe and more airports are affected (including AMS, freight gateways (AMS, CDG, FRA and LHR) affect the air CDG, FRA and LHR) and also the intensity of disruption freight system signifi cantly. has increased. Both scenarios constraint the air capacity of the system whereas hinterland connections to and from the airports keep unrestrained. Hence, a short- to medium-term 5. Conclusions perspective is applied in the present scenarios. The eruption of Iceland’s Eyjafjalljökull volcano eruption Detailed results of the scenarios are presented in Table in April 2010 impressively demonstrated the dependency 4 and Table 5. It becomes obvious that because of the of the economy on a smooth running air cargo network. interdependency and the interconnectedness of the air freight Though economic damages have been comparatively small, system, disruptions at selected airports (or regions) do impact production came to halt due to missing parts in different the entire European air freight system. Disruptions at 10% locations all over Europe. of the European airports where still 80% of freight tonnes can be handled (scenario 1) lead to an average reduction of There is a lively debate going on whether the closure of daily freight tonnes of only 2.02%. This result indicates that the airports has been justifi ed. While airlines are clearly in scenario 1 especially smaller airports are affected by the sceptical, public authorities point to the safety aspect. No disruption and that comprehensive airport alternatives exist. matter which side is right, the ash cloud illustrated that crises Contrarily, the impacts of scenario 2 (red) where the major strategies are hardly mature. European air freight gateways are directly affected (AMS, The presented paper aims to add a small mosaic to the ongoing CDG, FRA and LHR) lead to an overall reduction of air 9 Because the hinterland connections to and from the airports keep un- 8 It should be noted that the scenario “winter storm Europe” is only a fi c- restrained, the overall reductions at the airport (air side + hinterland tive scenario. The events of December 2010 only serve as a orientation. capacities) are smaller.

48 Der öff entliche Sektor - Th e Public Sector Vol. 37 (3-4) 2011 Implications of short-term Cargo Collapses on European Airports

Affected countries Austria, Czech Republic, Germany* Austria, Belgium, Czech Republic, Germany, (all airports) Luxembourg, the Netherlands, Swiss France: (BVA, CDG, LIL, MLH, ORY, SXB, Affected airports - XCR) (additional airports) Great Britain: (BOH, BRS, BZZ, EXT, GCI, JER, LCY, LGW, LHR, LTN, MSE, NQY, NWI, SOU, STN) * In sub-scenario 1 (left , light shade) a disruption share of 0.8 is assumed (80% of freight tons can still be handled at the airports). In sub-scenario 2 (right, dark shade) a disruption share of 0.2 is assumed.

Source: Vosen, 2011 Fig. 6. Regional disruption for the case-study “winter storm Europe”

Table 4. Overview of scenario results (aggregated level)

Scenario Affected Disruption share Average freight Average freight Change in countries at affected tonnes per day tonnes per day freight tonnes airports (without disaster) (with disaster) per day 1 (light AUT, CZE, 0.8 156,137 152,986 -2.02% shade) GER 2 (dark AUT, BEL, 0.2 156,137 128,626 -17.62% shade) CZE, FRA*, GBR*, GER, LUX, NED, SUI * Only partly (see Figure 6 for further details)

Source: authors’ own representation

discussion in this fi eld by setting up a European air cargo Since the table follows the main principles of the input- input-output table in the fi rst step. These tables reveal the output analysis, indicators of airports’ connectedness and strong interdependence of the air cargo market and present the importance can be estimated. Based on the Ghosh inverse airports’ absolute in- and outfl ows. In input-output notation, matrix we fi rst calculate forward multipliers which are the airport to airport relations defi ne the intermediate matrix. independent of absolute fl ows but give a good idea of the Flows from the corresponding hinterland to the airports are airports forward linkages. Not surprisingly the main hub the primary inputs whereas fl ows to the hinterland are the airports come up with rather large multipliers. In contrast, fi nal demands. airports at the end of the logistic chain generally show small

Vol. 37 (3-4) 2011 Der öff entliche Sektor - Th e Public Sector 49 A.B. Scholz, B. Mandel, A. Schaff er

Table 5. Top aff ected airports caused by the winter-storm scenarios according to their level of disruption Scenario 1 Scenario 2 Rank Airport Country Disruption Freight Rank Airport Country Disruption Freight (affected level tonnes (affected level tonnes -ness) before -ness) before disaster disaster [tonnes [tonnes per day] per day] 1 LEI* ESP 0.82 5.66 1 LEI ESP 0.30 5.66 2 KVA GRE 0.87 17.19 2 LSI GBR 0.33 21.24 3 CFU GRE 0.88 6.06 3 MJV ESP 0.35 11.99 4 FLR ITA 0.88 9.97 4 MME GBR 0.43 7.80 5 KTW POL 0.89 10.21 5 SVQ ESP 0.43 42.96 6 TSF ITA 0.90 41.30 6 TLN FRA 0.46 0.59 7 POZ POL 0.91 11.16 7 PUF FRA 0.47 6.56 8 GDN POL 0.91 10.14 8 KVA GRE 0.50 17.19 9 SJJ BIH 0.92 2.82 9 BES FRA 0.51 1.95 10 AOI ITA 0.93 4.40 10 TRF NOR 0.53 32.59 * LEI: Aeropuerto de Almeria (), KVA: Kavala „Alexander the Great“ (Greece), LSI: Sumburgh Airport (Great Britain), CFU: Corfu International Airport, „Ioannis Kapodistrias“ (Greece), MJV: Murcia-San Javier Airport (Spain), FLR: Florence Airport (Italy), MME: Durham Tees Valley Airport (Great Britain), KTW: Katowice International Airport (Poland), SVQ: (Spain), TSF: (Spain), TLN: Toulon-Hyères Airport (France), POZ: Poznań-Ławica Henryk Wieniawski Airport (Poland), PUF: Pau Pyrénées Airport (France), GDN: Gdansk Lech Walesa Airport (Poland), SJJ: Sarajevo International Airport (Bosnia and Herzegovina), BES: Brest Bretagne Airport (France), AOI: Ancona-Falconara Airport (Italy), TRF: Sandefj ord Airport (Norway)

Source: authors’ own representation

multipliers. The second step foresees the calculation of a The present steady-state approach weighted forward linkage. For this purpose two alternative  Elaborates the importance of the primary airports weights are suggested: the absolute cargo volume transported (gateway airports) for the entire air freight system to other airports and the logarithmic value of these fl ows. In case of the fi rst weight, the role of the connectedness is  Identifi es the high interconnectedness of the air freight reduced in favour of the airports size and the airports with system which is a risk (dependency from gateway the highest linkages follow very much the same ranking as if airports) but also an opportunity (alternative routing the ranking is purely based on the volumes. In contrast, the opportunities) application of the second weight seems to provide a linkage  Ascertains the vulnerability of the system in case of index with a sound balance of forward connectedness disruptions at gateway airports which is caused by the and absolute volumes. Interestingly the rankings differ sophisticated hub-and-spoke systems of the market signifi cantly compared to the fi rst alternative. In case a players (e.g. airlines, forwarders) crises strategy is not only driven by the airport size but takes into account the dependency as well the latter index can be considered a better proxy for the airports’ importance. In a last step an input-output model is developed which Acknowledgements quantifi es direct and indirect impacts of disasters for the air freight system. Therefore, interdependency and Die vorliegende Veröffentlichung entstand zum Teil im interconnectedness between the airports is considered. Rahmes des Forschungsprojektes RM-LOG (Risikoman- Hence, cascading effects can be represented by the model agementstrategien in Logistik- und Infrastrukturnetzen aus which is extremely important for network systems, such as unternehmerischer und gesamtwirtschaftlicher Sicht). RM- the air freight system where hub-and-spoke structures are the LOG wird im Zuge der Bekanntmachung “Sicherung der predominant network structures (Scholz, 2011). A case-study Warenketten” des Bundesministeriums für Bildung und application which is leant on the winter storm of December Forschung im Rahmen des Programms “Forschung für die 2010 shows the functionality of the model. zivile Sicherheit” der deutschen Bundesregierung gefördert.

50 Der öff entliche Sektor - Th e Public Sector Vol. 37 (3-4) 2011 Implications of short-term Cargo Collapses on European Airports

References Annex Dietzenbacher, E. (1997), In vindication of the Ghosh model: Calculation of the hinterland outfl ows for the artifi cial a reinterpretation as a price model, Journal of Regional rest of the world airports: Science, Volume 37, pp. 629-651.  Calculate the total tonnes for all European airports: Jiang, P. & Haimes, Y. Y. (2004), Risk management for 291 296 Leontief-based interdependent systems, Risk Analysis, 24(5), pp. 1215-1229. xsum  xij ij1 1 Nieuwenhuijs, A., Luiijf, E. & Klaver, M. (2008), Modelling dependencies in critical infrastructures, in Papa, M.  Calculate the hinterland fl ows (other outfl ows) for all Shenoi, S. (EDS.), Critical Infrastructure Protection II, European airports: Springer, Boston, Massachusetts, 2008, pp. 205-213. 291 e  e MergeGlobal (2006), Steady climb - MergeGlobal forecasts  i accelerating intercontinental air freight demand growth i1 through 2010, August 2006, Arlington, Virginia.  Calculate the European-Airport-Hinterland-Outfl ow- Schaffer, A. (2002), Ecological Input-Output Analysis, Coeffi cient: Nomos Verlagsgesellschaft Baden-Baden.

xsum Scholz, A.B. & von Cossel, J. (2011), Assessing the   importance of hub airports for cargo carriers and its e implications for a sustainable airport management,  Apply the European-Airport-Hinterland-Outfl ow- Research in Transportation Business & Management, Coeffi cient to the World Regions: Volume 1, Issue 1, August 2011, pp. 62-70.

296 Vázques, E.F., Muñiz, A.S.G., Carvajal, C.R. (2009), The impact of Immigration on interregional Migrations: An  xij j1 Input-Output Analysis with an application for Spain, ei  The Annals of Regional Science, Volume 46, Number  1, pp. 189-204, DOI: 10.1007/s00168-009-0331-6. The determination of the infl ows (hinterland) bases on Vosen, R. (2011), Auswirkungen von Katastrophenereignissen

the simulated outfl ows as the row sum of airport i (xi) auf die europäische Luftfracht, Diplomarbeit am

equals the column sum of airport i (xi). This assumption Institut für Wirtschaftspolitik und Wirtschaftsforschung can be justifi ed because the air freight market is a closed (IWW), Karlsruher Institut für Technologie (KIT), system that every tonne of freight must have one specifi c 2011. origin as well as one dedicated destination. The infl ows West. G.R. (1998), Structural change in the Queensland calculation passes the following procedure: economy: An Interindustry Analysis, Paper presented at 45th North American Meeting of the Regional  Calculate the total outfl ow tonnes (incl. hinterland) Science Association, Santa Fee, November 1998. for all airports (incl. World Regions): WORLDNET (2009a), Deliverable 7 of the WORLDNET 296 297 x  x Research Project: Freight Flows, download at: http:// i  ij www.WORLDNETproject.eu (last visited April, 20 ij1 1 2010).  Calculate the total infl ow tonnes (excl. hinterland) for WORLDNET (2009b), Deliverable 8 of the WORLDNET all airports (incl. World Regions): Research Project: Network Data, download at: http://

296 296 www.WORLDNETproject.eu (last visited April, 19 2010). xsum,i  xij ij1 1  Determine the missing hinterland infl ows for the World Regions:

i  x  x i i sum,i

Vol. 37 (3-4) 2011 Der öff entliche Sektor - Th e Public Sector 51